bgp.c

   1 /*
   2  * BGPv4
   3  * Used to advertise routes for upstream (l2tp port, rather than gratiutious
   4  * arp) and downstream--allowing routers to load-balance both.
   5  *
   6  * Implementation limitations:
   7  * - We never listen for incoming connections (session always initiated by us).
   8  * - Any routes advertised by the peer are accepted, but ignored.
   9  * - No password support; neither RFC1771 (which no-one seems to do anyway)
  10  *   nor RFC2385 (which requires a kernel patch on 2.4 kernels).
  11  */
  12
  13 char const *cvs_id_bgp = "$Id: bgp.c,v 1.12 2005/09/02 23:39:36 bodea Exp $";
  14
  15 #include <stdlib.h>
  16 #include <unistd.h>
  17 #include <string.h>
  18 #include <time.h>
  19 #include <errno.h>
  20 #include <sys/socket.h>
  21 #include <netinet/in.h>
  22 #include <arpa/inet.h>
  23 #include <netdb.h>
  24 #include <fcntl.h>
  25
  26 #include "l2tpns.h"
  27 #include "bgp.h"
  28 #include "util.h"
  29
  30 static void bgp_clear(struct bgp_peer *peer);
  31 static void bgp_set_retry(struct bgp_peer *peer);
  32 static void bgp_cidr(in_addr_t ip, in_addr_t mask, struct bgp_ip_prefix *pfx);
  33 static struct bgp_route_list *bgp_insert_route(struct bgp_route_list *head,
  34     struct bgp_route_list *new);
  35
  36 static void bgp_free_routes(struct bgp_route_list *routes);
  37 static char const *bgp_msg_type_str(uint8_t type);
  38 static int bgp_connect(struct bgp_peer *peer);
  39 static int bgp_handle_connect(struct bgp_peer *peer);
  40 static int bgp_write(struct bgp_peer *peer);
  41 static int bgp_read(struct bgp_peer *peer);
  42 static int bgp_handle_input(struct bgp_peer *peer);
  43 static int bgp_send_open(struct bgp_peer *peer);
  44 static int bgp_send_keepalive(struct bgp_peer *peer);
  45 static int bgp_send_update(struct bgp_peer *peer);
  46 static int bgp_send_notification(struct bgp_peer *peer, uint8_t code,
  47     uint8_t subcode);
  48 static int bgp_send_notification_full(struct bgp_peer *peer, uint8_t code,
  49     uint8_t subcode, char *notification_data, uint16_t data_len);
  50
  51 static uint16_t our_as;
  52 static struct bgp_route_list *bgp_routes = 0;
  53
  54 int bgp_configured = 0;
  55 struct bgp_peer *bgp_peers = 0;
  56
  57 /* prepare peer structure, globals */
  58 int bgp_setup(int as)
  59 {
  60     int i;
  61     struct bgp_peer *peer;
  62
  63     for (i = 0; i < BGP_NUM_PEERS; i++)
  64     {
  65         peer = &bgp_peers[i];
  66         memset(peer, 0, sizeof(*peer));
  67
  68         peer->addr = INADDR_NONE;
  69         peer->sock = -1;
  70         peer->state = peer->next_state = Disabled;
  71
  72         if (!((peer->outbuf = malloc(sizeof(*peer->outbuf)))
  73             && (peer->inbuf = malloc(sizeof(*peer->inbuf)))))
  74         {
  75             LOG(0, 0, 0, "Can't allocate buffers for bgp peer (%s)\n",
  76                 strerror(errno));
  77
  78             return 0;
  79         }
  80
  81         peer->edata.type = FD_TYPE_BGP;
  82         peer->edata.index = i;
  83         peer->events = 0;
  84     }
  85
  86     if (as < 1)
  87         as = 0;
  88
  89     if ((our_as = as))
  90         return 0;
  91
  92     bgp_routes = 0;
  93     bgp_configured = 0; /* set by bgp_start */
  94
  95     return 1;
  96 }
  97
  98 /* start connection with a peer */
  99 int bgp_start(struct bgp_peer *peer, char *name, int as, int keepalive,
 100     int hold, int enable)
 101 {
 102     struct hostent *h;
 103     int ibgp;
 104     int i;
 105     struct bgp_path_attr a;
 106     char path_attrs[64];
 107     char *p = path_attrs;
 108     in_addr_t ip;
 109     uint32_t metric = htonl(BGP_METRIC);
 110     uint32_t no_export = htonl(BGP_COMMUNITY_NO_EXPORT);
 111
 112     if (!our_as)
 113         return 0;
 114
 115     if (peer->state != Disabled)
 116         bgp_halt(peer);
 117
 118     snprintf(peer->name, sizeof(peer->name), "%s", name);
 119
 120     if (!(h = gethostbyname(name)) || h->h_addrtype != AF_INET)
 121     {
 122         LOG(0, 0, 0, "Can't get address for BGP peer %s (%s)\n",
 123             name, h ? "no address" : hstrerror(h_errno));
 124
 125         return 0;
 126     }
 127
 128     memcpy(&peer->addr, h->h_addr, sizeof(peer->addr));
 129     peer->as = as > 0 ? as : our_as;
 130     ibgp = peer->as == our_as;
 131
 132     /* set initial timer values */
 133     peer->init_keepalive = keepalive == -1 ? BGP_KEEPALIVE_TIME : keepalive;
 134     peer->init_hold = hold == -1 ? BGP_HOLD_TIME : hold;
 135
 136     if (peer->init_hold < 3)
 137         peer->init_hold = 3;
 138
 139     if (peer->init_keepalive * 3 > peer->init_hold)
 140         peer->init_keepalive = peer->init_hold / 3;
 141
 142     /* clear buffers, go to Idle state */
 143     peer->next_state = Idle;
 144     bgp_clear(peer);
 145
 146     /* set initial routing state */
 147     peer->routing = enable;
 148
 149     /* all our routes use the same attributes, so prepare it in advance */
 150     if (peer->path_attrs)
 151         free(peer->path_attrs);
 152
 153     peer->path_attr_len = 0;
 154
 155     /* ORIGIN */
 156     a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 157     a.code = BGP_PATH_ATTR_CODE_ORIGIN;
 158     a.data.s.len = 1;
 159     a.data.s.value[0] = BGP_PATH_ATTR_CODE_ORIGIN_IGP;
 160
 161 #define ADD_ATTRIBUTE()         do { \
 162     i = BGP_PATH_ATTR_SIZE(a);  \
 163     memcpy(p, &a, i);           \
 164     p += i;                     \
 165     peer->path_attr_len += i;   } while (0)
 166
 167     ADD_ATTRIBUTE();
 168
 169     /* AS_PATH */
 170     a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 171     a.code = BGP_PATH_ATTR_CODE_AS_PATH;
 172     if (ibgp)
 173     {
 174         /* empty path */
 175         a.data.s.len = 0;
 176     }
 177     else
 178     {
 179         /* just our AS */
 180         struct {
 181             uint8_t type;
 182             uint8_t len;
 183             uint16_t value;
 184         } as_path = {
 185             BGP_PATH_ATTR_CODE_AS_PATH_AS_SEQUENCE,
 186             1,
 187             htons(our_as),
 188         };
 189
 190         a.data.s.len = sizeof(as_path);
 191         memcpy(&a.data.s.value, &as_path, sizeof(as_path));
 192     }
 193
 194     ADD_ATTRIBUTE();
 195
 196     /* NEXT_HOP */
 197     a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 198     a.code = BGP_PATH_ATTR_CODE_NEXT_HOP;
 199     ip = my_address; /* we're it */
 200     a.data.s.len = sizeof(ip);
 201     memcpy(a.data.s.value, &ip, sizeof(ip));
 202
 203     ADD_ATTRIBUTE();
 204
 205     /* MULTI_EXIT_DISC */
 206     a.flags = BGP_PATH_ATTR_FLAG_OPTIONAL;
 207     a.code = BGP_PATH_ATTR_CODE_MULTI_EXIT_DISC;
 208     a.data.s.len = sizeof(metric);
 209     memcpy(a.data.s.value, &metric, sizeof(metric));
 210
 211     ADD_ATTRIBUTE();
 212
 213     if (ibgp)
 214     {
 215         uint32_t local_pref = htonl(BGP_LOCAL_PREF);
 216
 217         /* LOCAL_PREF */
 218         a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 219         a.code = BGP_PATH_ATTR_CODE_LOCAL_PREF;
 220         a.data.s.len = sizeof(local_pref);
 221         memcpy(a.data.s.value, &local_pref, sizeof(local_pref));
 222
 223         ADD_ATTRIBUTE();
 224     }
 225
 226     /* COMMUNITIES */
 227     a.flags = BGP_PATH_ATTR_FLAG_OPTIONAL | BGP_PATH_ATTR_FLAG_TRANS;
 228     a.code = BGP_PATH_ATTR_CODE_COMMUNITIES;
 229     a.data.s.len = sizeof(no_export);
 230     memcpy(a.data.s.value, &no_export, sizeof(no_export));
 231
 232     ADD_ATTRIBUTE();
 233
 234     if (!(peer->path_attrs = malloc(peer->path_attr_len)))
 235     {
 236         LOG(0, 0, 0, "Can't allocate path_attrs for %s (%s)\n",
 237             name, strerror(errno));
 238
 239         return 0;
 240     }
 241
 242     memcpy(peer->path_attrs, path_attrs, peer->path_attr_len);
 243
 244     LOG(4, 0, 0, "Initiating BGP connection to %s (routing %s)\n",
 245         name, enable ? "enabled" : "suspended");
 246
 247     /* we have at least one peer configured */
 248     bgp_configured = 1;
 249
 250     /* connect */
 251     return bgp_connect(peer);
 252 }
 253
 254 /* clear counters, timers, routes and buffers; close socket; move to
 255    next_state, which may be Disabled or Idle */
 256 static void bgp_clear(struct bgp_peer *peer)
 257 {
 258     if (peer->sock != -1)
 259     {
 260         close(peer->sock);
 261         peer->sock = -1;
 262     }
 263
 264     peer->keepalive_time = 0;
 265     peer->expire_time = 0;
 266
 267     peer->keepalive = peer->init_keepalive;
 268     peer->hold = peer->init_hold;
 269
 270     bgp_free_routes(peer->routes);
 271     peer->routes = 0;
 272
 273     peer->outbuf->packet.header.len = 0;
 274     peer->outbuf->done = 0;
 275     peer->inbuf->packet.header.len = 0;
 276     peer->inbuf->done = 0;
 277
 278     peer->cli_flag = 0;
 279     peer->events = 0;
 280
 281     if (peer->state != peer->next_state)
 282     {
 283         peer->state = peer->next_state;
 284         peer->state_time = time_now;
 285
 286         LOG(4, 0, 0, "BGP peer %s: state %s\n", peer->name,
 287             bgp_state_str(peer->next_state));
 288     }
 289 }
 290
 291 /* initiate a clean shutdown */
 292 void bgp_stop(struct bgp_peer *peer)
 293 {
 294     LOG(4, 0, 0, "Terminating BGP connection to %s\n", peer->name);
 295     bgp_send_notification(peer, BGP_ERR_CEASE, 0);
 296 }
 297
 298 /* drop connection (if any) and set state to Disabled */
 299 void bgp_halt(struct bgp_peer *peer)
 300 {
 301     LOG(4, 0, 0, "Aborting BGP connection to %s\n", peer->name);
 302     peer->next_state = Disabled;
 303     bgp_clear(peer);
 304 }
 305
 306 /* drop connection (if any) and set to Idle for connection retry */
 307 int bgp_restart(struct bgp_peer *peer)
 308 {
 309     peer->next_state = Idle;
 310     bgp_clear(peer);
 311
 312     /* restart now */
 313     peer->retry_time = time_now;
 314     peer->retry_count = 0;
 315
 316     /* connect */
 317     return bgp_connect(peer);
 318 }
 319
 320 static void bgp_set_retry(struct bgp_peer *peer)
 321 {
 322     if (peer->retry_count++ < BGP_MAX_RETRY)
 323     {
 324         peer->retry_time = time_now + (BGP_RETRY_BACKOFF * peer->retry_count);
 325         peer->next_state = Idle;
 326         bgp_clear(peer);
 327     }
 328     else
 329         bgp_halt(peer); /* give up */
 330 }
 331
 332 /* convert ip/mask to CIDR notation */
 333 static void bgp_cidr(in_addr_t ip, in_addr_t mask, struct bgp_ip_prefix *pfx)
 334 {
 335     int i;
 336     uint32_t b;
 337
 338     /* convert to prefix notation */
 339     pfx->len = 32;
 340     pfx->prefix = ip;
 341
 342     if (!mask) /* bogus */
 343         mask = 0xffffffff;
 344
 345     for (i = 0; i < 32 && ((b = ntohl(1 << i)), !(mask & b)); i++)
 346     {
 347         pfx->len--;
 348         pfx->prefix &= ~b;
 349     }
 350 }
 351
 352 /* insert route into list; sorted */
 353 static struct bgp_route_list *bgp_insert_route(struct bgp_route_list *head,
 354     struct bgp_route_list *new)
 355 {
 356     struct bgp_route_list *p = head;
 357     struct bgp_route_list *e = 0;
 358
 359     while (p && memcmp(&p->dest, &new->dest, sizeof(p->dest)) < 0)
 360     {
 361         e = p;
 362         p = p->next;
 363     }
 364
 365     if (e)
 366     {
 367         new->next = e->next;
 368         e->next = new;
 369     }
 370     else
 371     {
 372         new->next = head;
 373         head = new;
 374     }
 375
 376     return head;
 377 }
 378
 379 /* add route to list for peers */
 380 /*
 381  * Note:  this doesn't do route aggregation, nor drop routes if a less
 382  * specific match already exists (partly because I'm lazy, but also so
 383  * that if that route is later deleted we don't have to be concerned
 384  * about adding back the more specific one).
 385  */
 386 int bgp_add_route(in_addr_t ip, in_addr_t mask)
 387 {
 388     struct bgp_route_list *r = bgp_routes;
 389     struct bgp_route_list add;
 390     int i;
 391
 392     bgp_cidr(ip, mask, &add.dest);
 393     add.next = 0;
 394
 395     /* check for duplicate */
 396     while (r)
 397     {
 398         i = memcmp(&r->dest, &add.dest, sizeof(r->dest));
 399         if (!i)
 400             return 1; /* already covered */
 401
 402         if (i > 0)
 403             break;
 404
 405         r = r->next;
 406     }
 407
 408     /* insert into route list; sorted */
 409     if (!(r = malloc(sizeof(*r))))
 410     {
 411         LOG(0, 0, 0, "Can't allocate route for %s/%d (%s)\n",
 412             fmtaddr(add.dest.prefix, 0), add.dest.len, strerror(errno));
 413
 414         return 0;
 415     }
 416
 417     memcpy(r, &add, sizeof(*r));
 418     bgp_routes = bgp_insert_route(bgp_routes, r);
 419
 420     /* flag established peers for update */
 421     for (i = 0; i < BGP_NUM_PEERS; i++)
 422         if (bgp_peers[i].state == Established)
 423             bgp_peers[i].update_routes = 1;
 424
 425     LOG(4, 0, 0, "Registered BGP route %s/%d\n",
 426         fmtaddr(add.dest.prefix, 0), add.dest.len);
 427
 428     return 1;
 429 }
 430
 431 /* remove route from list for peers */
 432 int bgp_del_route(in_addr_t ip, in_addr_t mask)
 433 {
 434     struct bgp_route_list *r = bgp_routes;
 435     struct bgp_route_list *e = 0;
 436     struct bgp_route_list del;
 437     int i;
 438
 439     bgp_cidr(ip, mask, &del.dest);
 440     del.next = 0;
 441
 442     /* find entry in routes list and remove */
 443     while (r)
 444     {
 445         i = memcmp(&r->dest, &del.dest, sizeof(r->dest));
 446         if (!i)
 447         {
 448             if (e)
 449                 e->next = r->next;
 450             else
 451                 bgp_routes = r->next;
 452
 453             free(r);
 454             break;
 455         }
 456
 457         e = r;
 458
 459         if (i > 0)
 460             r = 0; /* stop */
 461         else
 462             r = r->next;
 463     }
 464
 465     /* not found */
 466     if (!r)
 467         return 1;
 468
 469     /* flag established peers for update */
 470     for (i = 0; i < BGP_NUM_PEERS; i++)
 471         if (bgp_peers[i].state == Established)
 472             bgp_peers[i].update_routes = 1;
 473
 474     LOG(4, 0, 0, "Removed BGP route %s/%d\n",
 475         fmtaddr(del.dest.prefix, 0), del.dest.len);
 476
 477     return 1;
 478 }
 479
 480 /* enable or disable routing */
 481 void bgp_enable_routing(int enable)
 482 {
 483     int i;
 484
 485     for (i = 0; i < BGP_NUM_PEERS; i++)
 486     {
 487         bgp_peers[i].routing = enable;
 488
 489         /* flag established peers for update */
 490         if (bgp_peers[i].state == Established)
 491             bgp_peers[i].update_routes = 1;
 492     }
 493
 494     LOG(4, 0, 0, "%s BGP routing\n", enable ? "Enabled" : "Suspended");
 495 }
 496
 497 #ifdef HAVE_EPOLL
 498 # include <sys/epoll.h>
 499 #else
 500 # include "fake_epoll.h"
 501 #endif
 502
 503 /* return a bitmask of the events required to poll this peer's fd */
 504 int bgp_set_poll()
 505 {
 506     int i;
 507
 508     if (!bgp_configured)
 509         return 0;
 510
 511     for (i = 0; i < BGP_NUM_PEERS; i++)
 512     {
 513         struct bgp_peer *peer = &bgp_peers[i];
 514         int events = 0;
 515
 516         if (peer->state == Disabled || peer->state == Idle)
 517             continue;
 518
 519         if (peer->inbuf->done < BGP_MAX_PACKET_SIZE)
 520             events |= EPOLLIN;
 521
 522         if (peer->state == Connect ||           /* connection in progress */
 523             peer->update_routes ||              /* routing updates */
 524             peer->outbuf->packet.header.len)    /* pending output */
 525             events |= EPOLLOUT;
 526
 527         if (peer->events != events)
 528         {
 529             struct epoll_event ev;
 530
 531             ev.events = peer->events = events;
 532             ev.data.ptr = &peer->edata;
 533             epoll_ctl(epollfd, EPOLL_CTL_MOD, peer->sock, &ev);
 534         }
 535     }
 536
 537     return 1;
 538 }
 539
 540 /* process bgp events/timers */
 541 int bgp_process(uint32_t events[])
 542 {
 543     int i;
 544
 545     if (!bgp_configured)
 546         return 0;
 547
 548     for (i = 0; i < BGP_NUM_PEERS; i++)
 549     {
 550         struct bgp_peer *peer = &bgp_peers[i];
 551
 552         if (*peer->name && peer->cli_flag == BGP_CLI_RESTART)
 553         {
 554             bgp_restart(peer);
 555             continue;
 556         }
 557
 558         if (peer->state == Disabled)
 559             continue;
 560
 561         if (peer->cli_flag)
 562         {
 563             switch (peer->cli_flag)
 564             {
 565             case BGP_CLI_SUSPEND:
 566                 if (peer->routing)
 567                 {
 568                     peer->routing = 0;
 569                     if (peer->state == Established)
 570                         peer->update_routes = 1;
 571                 }
 572
 573                 break;
 574
 575             case BGP_CLI_ENABLE:
 576                 if (!peer->routing)
 577                 {
 578                     peer->routing = 1;
 579                     if (peer->state == Established)
 580                         peer->update_routes = 1;
 581                 }
 582
 583                 break;
 584             }
 585
 586             peer->cli_flag = 0;
 587         }
 588
 589         /* handle empty/fill of buffers */
 590         if (events[i] & EPOLLOUT)
 591         {
 592             int r = 1;
 593             if (peer->state == Connect)
 594                 r = bgp_handle_connect(peer);
 595             else if (peer->outbuf->packet.header.len)
 596                 r = bgp_write(peer);
 597
 598             if (!r)
 599                 continue;
 600         }
 601
 602         if (events[i] & (EPOLLIN|EPOLLHUP))
 603         {
 604             if (!bgp_read(peer))
 605                 continue;
 606         }
 607
 608         /* process input buffer contents */
 609         while (peer->inbuf->done >= sizeof(peer->inbuf->packet.header)
 610             && !peer->outbuf->packet.header.len) /* may need to queue a response */
 611         {
 612             if (bgp_handle_input(peer) < 0)
 613                 continue;
 614         }
 615
 616         /* process pending updates */
 617         if (peer->update_routes
 618             && !peer->outbuf->packet.header.len) /* ditto */
 619         {
 620             if (!bgp_send_update(peer))
 621                 continue;
 622         }
 623
 624         /* process timers */
 625         if (peer->state == Established)
 626         {
 627             if (time_now > peer->expire_time)
 628             {
 629                 LOG(1, 0, 0, "No message from BGP peer %s in %ds\n",
 630                     peer->name, peer->hold);
 631
 632                 bgp_send_notification(peer, BGP_ERR_HOLD_TIMER_EXP, 0);
 633                 continue;
 634             }
 635
 636             if (time_now > peer->keepalive_time && !peer->outbuf->packet.header.len)
 637                 bgp_send_keepalive(peer);
 638         }
 639         else if (peer->state == Idle)
 640         {
 641             if (time_now > peer->retry_time)
 642                 bgp_connect(peer);
 643         }
 644         else if (time_now > peer->state_time + BGP_STATE_TIME)
 645         {
 646             LOG(1, 0, 0, "%s timer expired for BGP peer %s\n",
 647                 bgp_state_str(peer->state), peer->name);
 648
 649             bgp_restart(peer);
 650         }
 651     }
 652
 653     return 1;
 654 }
 655
 656 static void bgp_free_routes(struct bgp_route_list *routes)
 657 {
 658     struct bgp_route_list *tmp;
 659
 660     while ((tmp = routes))
 661     {
 662         routes = tmp->next;
 663         free(tmp);
 664     }
 665 }
 666
 667 char const *bgp_state_str(enum bgp_state state)
 668 {
 669     switch (state)
 670     {
 671     case Disabled:      return "Disabled";
 672     case Idle:          return "Idle";
 673     case Connect:       return "Connect";
 674     case Active:        return "Active";
 675     case OpenSent:      return "OpenSent";
 676     case OpenConfirm:   return "OpenConfirm";
 677     case Established:   return "Established";
 678     }
 679
 680     return "?";
 681 }
 682
 683 static char const *bgp_msg_type_str(uint8_t type)
 684 {
 685     switch (type)
 686     {
 687     case BGP_MSG_OPEN:          return "OPEN";
 688     case BGP_MSG_UPDATE:        return "UPDATE";
 689     case BGP_MSG_NOTIFICATION:  return "NOTIFICATION";
 690     case BGP_MSG_KEEPALIVE:     return "KEEPALIVE";
 691     }
 692
 693     return "?";
 694 }
 695
 696 /* attempt to connect to peer */
 697 static int bgp_connect(struct bgp_peer *peer)
 698 {
 699     static int bgp_port = 0;
 700     struct sockaddr_in addr;
 701     struct epoll_event ev;
 702
 703     if (!bgp_port)
 704     {
 705         struct servent *serv;
 706         if (!(serv = getservbyname("bgp", "tcp")))
 707         {
 708             LOG(0, 0, 0, "Can't get bgp service (%s)\n", strerror(errno));
 709             return 0;
 710         }
 711
 712         bgp_port = serv->s_port;
 713     }
 714
 715     if ((peer->sock = socket(PF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
 716     {
 717         LOG(0, 0, 0, "Can't create a socket for BGP peer %s (%s)\n",
 718             peer->name, strerror(errno));
 719
 720         peer->state = peer->next_state = Disabled;
 721         return 0;
 722     }
 723
 724     /* add to poll set */
 725     ev.events = peer->events = EPOLLOUT;
 726     ev.data.ptr = &peer->edata;
 727     epoll_ctl(epollfd, EPOLL_CTL_ADD, peer->sock, &ev);
 728
 729     /* set to non-blocking */
 730     fcntl(peer->sock, F_SETFL, fcntl(peer->sock, F_GETFL, 0) | O_NONBLOCK);
 731
 732     /* try connect */
 733     memset(&addr, 0, sizeof(addr));
 734     addr.sin_family = AF_INET;
 735     addr.sin_port = bgp_port;
 736     addr.sin_addr.s_addr = peer->addr;
 737
 738     while (connect(peer->sock, (struct sockaddr *) &addr, sizeof(addr)) == -1)
 739     {
 740         if (errno == EINTR) /* SIGALARM handler */
 741             continue;
 742
 743         if (errno != EINPROGRESS)
 744         {
 745             LOG(1, 0, 0, "Can't connect to BGP peer %s (%s)\n",
 746                 inet_ntoa(addr.sin_addr), strerror(errno));
 747
 748             bgp_set_retry(peer);
 749             return 0;
 750         }
 751
 752         peer->state = Connect;
 753         peer->state_time = time_now;
 754
 755         LOG(4, 0, 0, "BGP peer %s: state Connect\n", peer->name);
 756         return 1;
 757     }
 758
 759     peer->state = Active;
 760     peer->state_time = time_now;
 761     peer->retry_time = peer->retry_count = 0;
 762
 763     LOG(4, 0, 0, "BGP peer %s: state Active\n", inet_ntoa(addr.sin_addr));
 764
 765     return bgp_send_open(peer);
 766 }
 767
 768 /* complete partial connection (state = Connect) */
 769 static int bgp_handle_connect(struct bgp_peer *peer)
 770 {
 771     int err = 0;
 772     socklen_t len = sizeof(int);
 773     getsockopt(peer->sock, SOL_SOCKET, SO_ERROR, &err, &len);
 774     if (err)
 775     {
 776         LOG(1, 0, 0, "Can't connect to BGP peer %s (%s)\n", peer->name,
 777             strerror(err));
 778
 779         bgp_set_retry(peer);
 780         return 0;
 781     }
 782
 783     peer->state = Active;
 784     peer->state_time = time_now;
 785
 786     LOG(4, 0, 0, "BGP peer %s: state Active\n", peer->name);
 787
 788     return bgp_send_open(peer);
 789 }
 790
 791 /* initiate a write */
 792 static int bgp_write(struct bgp_peer *peer)
 793 {
 794     int len = htons(peer->outbuf->packet.header.len);
 795     int r;
 796
 797     while ((r = write(peer->sock, &peer->outbuf->packet + peer->outbuf->done,
 798         len - peer->outbuf->done)) == -1)
 799     {
 800         if (errno == EINTR)
 801             continue;
 802
 803         if (errno == EAGAIN)
 804             return 1;
 805
 806         if (errno == EPIPE)
 807             LOG(1, 0, 0, "Connection to BGP peer %s closed\n", peer->name);
 808         else
 809             LOG(1, 0, 0, "Can't write to BGP peer %s (%s)\n", peer->name,
 810                 strerror(errno));
 811
 812         bgp_set_retry(peer);
 813         return 0;
 814     }
 815
 816     if (r < len)
 817     {
 818         peer->outbuf->done += r;
 819         return 1;
 820     }
 821
 822     LOG(4, 0, 0, "Sent %s to BGP peer %s\n",
 823         bgp_msg_type_str(peer->outbuf->packet.header.type), peer->name);
 824
 825     peer->outbuf->packet.header.len = 0;
 826     peer->outbuf->done = 0;
 827
 828     if (peer->state == Established)
 829         peer->keepalive_time = time_now + peer->keepalive;
 830
 831     if (peer->state != peer->next_state)
 832     {
 833         if (peer->next_state == Disabled || peer->next_state == Idle)
 834         {
 835             bgp_clear(peer);
 836             return 0;
 837         }
 838
 839         peer->state = peer->next_state;
 840         peer->state_time = time_now;
 841
 842         LOG(4, 0, 0, "BGP peer %s: state %s\n", peer->name,
 843             bgp_state_str(peer->state));
 844     }
 845
 846     return 1;
 847 }
 848
 849 /* initiate a read */
 850 static int bgp_read(struct bgp_peer *peer)
 851 {
 852     int r;
 853
 854     while ((r = read(peer->sock, &peer->inbuf->packet + peer->inbuf->done,
 855         BGP_MAX_PACKET_SIZE - peer->inbuf->done)) < 1)
 856     {
 857         if (!r)
 858         {
 859             LOG(1, 0, 0, "Connection to BGP peer %s closed\n", peer->name);
 860         }
 861         else
 862         {
 863             if (errno == EINTR)
 864                 continue;
 865
 866             if (errno == EAGAIN)
 867                 return 1;
 868
 869             LOG(1, 0, 0, "Can't read from BGP peer %s (%s)\n", peer->name,
 870                 strerror(errno));
 871         }
 872
 873         bgp_set_retry(peer);
 874         return 0;
 875     }
 876
 877     peer->inbuf->done += r;
 878     return 1;
 879 }
 880
 881 /* process buffered packets */
 882 static int bgp_handle_input(struct bgp_peer *peer)
 883 {
 884     struct bgp_packet *p = &peer->inbuf->packet;
 885     int len = ntohs(p->header.len);
 886
 887     if (len > BGP_MAX_PACKET_SIZE)
 888     {
 889         LOG(1, 0, 0, "Bad header length from BGP %s\n", peer->name);
 890         bgp_send_notification(peer, BGP_ERR_HEADER, BGP_ERR_HDR_BAD_LEN);
 891         return 0;
 892     }
 893
 894     if (peer->inbuf->done < len)
 895         return 0;
 896
 897     LOG(4, 0, 0, "Received %s from BGP peer %s\n",
 898         bgp_msg_type_str(p->header.type), peer->name);
 899
 900     switch (p->header.type)
 901     {
 902     case BGP_MSG_OPEN:
 903         {
 904             struct bgp_data_open data;
 905             int hold;
 906             int i;
 907             off_t param_offset, capability_offset;
 908             struct bgp_opt_param *param;
 909             uint8_t capabilities_len;
 910             char *capabilities = NULL;
 911             struct bgp_capability *capability;
 912             struct bgp_mp_cap_param *mp_cap;
 913
 914             for (i = 0; i < sizeof(p->header.marker); i++)
 915             {
 916                 if ((unsigned char) p->header.marker[i] != 0xff)
 917                 {
 918                     LOG(1, 0, 0, "Invalid marker from BGP peer %s\n",
 919                         peer->name);
 920
 921                     bgp_send_notification(peer, BGP_ERR_HEADER,
 922                         BGP_ERR_HDR_NOT_SYNC);
 923
 924                     return 0;
 925                 }
 926             }
 927
 928             if (peer->state != OpenSent)
 929             {
 930                 LOG(1, 0, 0, "OPEN from BGP peer %s in %s state\n",
 931                     peer->name, bgp_state_str(peer->state));
 932
 933                 bgp_send_notification(peer, BGP_ERR_FSM, 0);
 934                 return 0;
 935             }
 936
 937             memcpy(&data, p->data, len - sizeof(p->header));
 938
 939             if (data.version != BGP_VERSION)
 940             {
 941                 LOG(1, 0, 0, "Bad version (%d) sent by BGP peer %s\n",
 942                     (int) data.version, peer->name);
 943
 944                 bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_VERSION);
 945                 return 0;
 946             }
 947
 948             if (ntohs(data.as) != peer->as)
 949             {
 950                 LOG(1, 0, 0, "Bad AS sent by BGP peer %s (got %d, "
 951                     "expected %d)\n", peer->name, (int) htons(data.as),
 952                     (int) peer->as);
 953
 954                 bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_BAD_AS);
 955                 return 0;
 956             }
 957
 958             if ((hold = ntohs(data.hold_time)) < 3)
 959             {
 960                 LOG(1, 0, 0, "Bad hold time (%d) from BGP peer %s\n",
 961                     hold, peer->name);
 962
 963                 bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_HOLD_TIME);
 964                 return 0;
 965             }
 966
 967             /* pick lowest hold time */
 968             if (hold < peer->hold)
 969                 peer->hold = hold;
 970
 971             /* adjust our keepalive based on negotiated hold value */
 972             if (peer->keepalive * 3 > peer->hold)
 973                 peer->keepalive = peer->hold / 3;
 974
 975             /* check for optional parameters */
 976             /* 2 is for the size of type + len (both uint8_t) */
 977             for (param_offset = 0;
 978                     param_offset < data.opt_len;
 979                     param_offset += 2 + param->len)
 980             {
 981                 param = (struct bgp_opt_param *)(&data.opt_params + param_offset);
 982
 983                 /* sensible check */
 984                 if (data.opt_len - param_offset < 2
 985                         || param->len > data.opt_len - param_offset - 2)
 986                 {
 987                     LOG(1, 0, 0, "Malformed Optional Parameter list from BGP peer %s\n",
 988                         peer->name);
 989
 990                     bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_UNSPEC);
 991                     return 0;
 992                 }
 993
 994                 /* we know only one parameter type */
 995                 if (param->type != BGP_CAPABILITY_PARAM_TYPE)
 996                 {
 997                     LOG(1, 0, 0, "Unsupported Optional Parameter type %d from BGP peer %s\n",
 998                         param->type, peer->name);
 999
1000                     bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_UNSUP_PARAM);
1001                     return 0;
1002                 }
1003
1004                 capabilities_len = param->len;
1005                 capabilities = (char *)&param->value;
1006             }
1007
1008             /* look for BGP multiprotocol capability */
1009             if (capabilities)
1010             {
1011                 for (capability_offset = 0;
1012                         capability_offset < capabilities_len;
1013                         capability_offset += 2 + capability->len)
1014                 {
1015                     capability = (struct bgp_capability *)(capabilities + capability_offset);
1016
1017                     /* sensible check */
1018                     if (capabilities_len - capability_offset < 2
1019                             || capability->len > capabilities_len - capability_offset - 2)
1020                     {
1021                         LOG(1, 0, 0, "Malformed Capabilities list from BGP peer %s\n",
1022                             peer->name);
1023
1024                         bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_UNSPEC);
1025                         return 0;
1026                     }
1027
1028                     /* we only know one capability code */
1029                     if (capability->code != BGP_CAP_CODE_MP
1030                             && capability->len != sizeof(struct bgp_mp_cap_param))
1031                     {
1032                         LOG(4, 0, 0, "Unsupported Capability code %d from BGP peer %s\n",
1033                             capability->code, peer->name);
1034
1035                         bgp_send_notification_full(peer, BGP_ERR_OPEN, BGP_ERR_OPN_UNSUP_CAP,
1036                                 capability, 2 + capability->len);
1037                         /* we don't terminate, still; we just jump to the next one */
1038                         continue;
1039                     }
1040
1041                     mp_cap = (struct bgp_mp_cap_param *)&capability->value;
1042                     /* the only <AFI, SAFI> tuple we support */
1043                     if (mp_cap->afi != AF_INET6 && mp_cap->safi != BGP_MP_SAFI_UNICAST)
1044                     {
1045                         LOG(4, 0, 0, "Unsupported multiprotocol AFI %d and SAFI %d from BGP peer %s\n",
1046                             mp_cap->afi, mp_cap->safi, peer->name);
1047
1048                         bgp_send_notification_full(peer, BGP_ERR_OPEN, BGP_ERR_OPN_UNSUP_CAP,
1049                                 capability, 2 + capability->len);
1050                         /* we don't terminate, still; we just jump to the next one */
1051                         continue;
1052                     }
1053
1054                 }
1055             }
1056
1057             /* next transition requires an exchange of keepalives */
1058             bgp_send_keepalive(peer);
1059         }
1060
1061         break;
1062
1063     case BGP_MSG_KEEPALIVE:
1064         if (peer->state == OpenConfirm)
1065         {
1066             peer->state = peer->next_state = Established;
1067             peer->state_time = time_now;
1068             peer->keepalive_time = time_now + peer->keepalive;
1069             peer->update_routes = 1;
1070             peer->retry_count = 0;
1071             peer->retry_time = 0;
1072
1073             LOG(4, 0, 0, "BGP peer %s: state Established\n", peer->name);
1074         }
1075
1076         break;
1077
1078     case BGP_MSG_NOTIFICATION:
1079         if (len > sizeof(p->header))
1080         {
1081             struct bgp_data_notification *notification =
1082                 (struct bgp_data_notification *) p->data;
1083
1084             if (notification->error_code == BGP_ERR_CEASE)
1085             {
1086                 LOG(4, 0, 0, "BGP peer %s sent CEASE\n", peer->name);
1087                 bgp_restart(peer);
1088                 return 0;
1089             }
1090
1091             /* FIXME: should handle more notifications */
1092             LOG(4, 0, 0, "BGP peer %s sent unhandled NOTIFICATION %d\n",
1093                 peer->name, (int) notification->error_code);
1094         }
1095
1096         break;
1097     }
1098
1099     /* reset timer */
1100     peer->expire_time = time_now + peer->hold;
1101
1102     /* see if there's another message in the same packet/buffer */
1103     if (peer->inbuf->done > len)
1104     {
1105         peer->inbuf->done -= len;
1106         memmove(p, (char *) p + len, peer->inbuf->done);
1107     }
1108     else
1109     {
1110         peer->inbuf->packet.header.len = 0;
1111         peer->inbuf->done = 0;
1112     }
1113
1114     return peer->inbuf->done;
1115 }
1116
1117 /* send/buffer OPEN message */
1118 static int bgp_send_open(struct bgp_peer *peer)
1119 {
1120     struct bgp_data_open data;
1121     uint16_t len = sizeof(peer->outbuf->packet.header);
1122
1123     memset(peer->outbuf->packet.header.marker, 0xff,
1124         sizeof(peer->outbuf->packet.header.marker));
1125
1126     peer->outbuf->packet.header.type = BGP_MSG_OPEN;
1127
1128     data.version = BGP_VERSION;
1129     data.as = htons(our_as);
1130     data.hold_time = htons(peer->hold);
1131     data.identifier = my_address;
1132     data.opt_len = 0;
1133
1134     memcpy(peer->outbuf->packet.data, &data, BGP_DATA_OPEN_SIZE);
1135     len += BGP_DATA_OPEN_SIZE;
1136
1137     peer->outbuf->packet.header.len = htons(len);
1138     peer->outbuf->done = 0;
1139     peer->next_state = OpenSent;
1140
1141     return bgp_write(peer);
1142 }
1143
1144 /* send/buffer KEEPALIVE message */
1145 static int bgp_send_keepalive(struct bgp_peer *peer)
1146 {
1147     memset(peer->outbuf->packet.header.marker, 0xff,
1148         sizeof(peer->outbuf->packet.header.marker));
1149
1150     peer->outbuf->packet.header.type = BGP_MSG_KEEPALIVE;
1151     peer->outbuf->packet.header.len =
1152         htons(sizeof(peer->outbuf->packet.header));
1153
1154     peer->outbuf->done = 0;
1155     peer->next_state = (peer->state == OpenSent) ? OpenConfirm : peer->state;
1156
1157     return bgp_write(peer);
1158 }
1159
1160 /* send/buffer UPDATE message */
1161 static int bgp_send_update(struct bgp_peer *peer)
1162 {
1163     uint16_t unf_len = 0;
1164     uint16_t attr_len;
1165     uint16_t len = sizeof(peer->outbuf->packet.header);
1166     struct bgp_route_list *have = peer->routes;
1167     struct bgp_route_list *want = peer->routing ? bgp_routes : 0;
1168     struct bgp_route_list *e = 0;
1169     struct bgp_route_list *add = 0;
1170     int s;
1171
1172     char *data = (char *) &peer->outbuf->packet.data;
1173
1174     /* need leave room for attr_len, bgp_path_attrs and one prefix */
1175     char *max = (char *) &peer->outbuf->packet.data
1176         + sizeof(peer->outbuf->packet.data)
1177         - sizeof(attr_len) - peer->path_attr_len - sizeof(struct bgp_ip_prefix);
1178
1179     /* skip over unf_len */
1180     data += sizeof(unf_len);
1181     len += sizeof(unf_len);
1182
1183     memset(peer->outbuf->packet.header.marker, 0xff,
1184         sizeof(peer->outbuf->packet.header.marker));
1185
1186     peer->outbuf->packet.header.type = BGP_MSG_UPDATE;
1187
1188     peer->update_routes = 0; /* tentatively clear */
1189
1190     /* find differences */
1191     while ((have || want) && data < (max - sizeof(struct bgp_ip_prefix)))
1192     {
1193         if (have)
1194             s = want
1195                 ? memcmp(&have->dest, &want->dest, sizeof(have->dest))
1196                 : -1;
1197         else
1198             s = 1;
1199
1200         if (s < 0) /* found one to delete */
1201         {
1202             struct bgp_route_list *tmp = have;
1203             have = have->next;
1204
1205             s = BGP_IP_PREFIX_SIZE(tmp->dest);
1206             memcpy(data, &tmp->dest, s);
1207             data += s;
1208             unf_len += s;
1209             len += s;
1210
1211             LOG(5, 0, 0, "Withdrawing route %s/%d from BGP peer %s\n",
1212                 fmtaddr(tmp->dest.prefix, 0), tmp->dest.len, peer->name);
1213
1214             free(tmp);
1215
1216             if (e)
1217                 e->next = have;
1218             else
1219                 peer->routes = have;
1220         }
1221         else
1222         {
1223             if (!s) /* same */
1224             {
1225                 e = have; /* stash the last found to relink above */
1226                 have = have->next;
1227                 want = want->next;
1228             }
1229             else if (s > 0) /* addition reqd. */
1230             {
1231                 if (add)
1232                 {
1233                     peer->update_routes = 1; /* only one add per packet */
1234                     if (!have)
1235                         break;
1236                 }
1237                 else
1238                     add = want;
1239
1240                 if (want)
1241                     want = want->next;
1242             }
1243         }
1244     }
1245
1246     if (have || want)
1247         peer->update_routes = 1; /* more to do */
1248
1249     /* anything changed? */
1250     if (!(unf_len || add))
1251         return 1;
1252
1253     /* go back and insert unf_len */
1254     unf_len = htons(unf_len);
1255     memcpy(&peer->outbuf->packet.data, &unf_len, sizeof(unf_len));
1256
1257     if (add)
1258     {
1259         if (!(e = malloc(sizeof(*e))))
1260         {
1261             LOG(0, 0, 0, "Can't allocate route for %s/%d (%s)\n",
1262                 fmtaddr(add->dest.prefix, 0), add->dest.len, strerror(errno));
1263
1264             return 0;
1265         }
1266
1267         memcpy(e, add, sizeof(*e));
1268         e->next = 0;
1269         peer->routes = bgp_insert_route(peer->routes, e);
1270
1271         attr_len = htons(peer->path_attr_len);
1272         memcpy(data, &attr_len, sizeof(attr_len));
1273         data += sizeof(attr_len);
1274         len += sizeof(attr_len);
1275
1276         memcpy(data, peer->path_attrs, peer->path_attr_len);
1277         data += peer->path_attr_len;
1278         len += peer->path_attr_len;
1279
1280         s = BGP_IP_PREFIX_SIZE(add->dest);
1281         memcpy(data, &add->dest, s);
1282         data += s;
1283         len += s;
1284
1285         LOG(5, 0, 0, "Advertising route %s/%d to BGP peer %s\n",
1286             fmtaddr(add->dest.prefix, 0), add->dest.len, peer->name);
1287     }
1288     else
1289     {
1290         attr_len = 0;
1291         memcpy(data, &attr_len, sizeof(attr_len));
1292         data += sizeof(attr_len);
1293         len += sizeof(attr_len);
1294     }
1295
1296     peer->outbuf->packet.header.len = htons(len);
1297     peer->outbuf->done = 0;
1298
1299     return bgp_write(peer);
1300 }
1301
1302 /* send/buffer NOTIFICATION message */
1303 static int bgp_send_notification(struct bgp_peer *peer, uint8_t code,
1304     uint8_t subcode)
1305 {
1306     return bgp_send_notification_full(peer, code, subcode, NULL, 0);
1307 }
1308
1309 static int bgp_send_notification_full(struct bgp_peer *peer, uint8_t code,
1310     uint8_t subcode, char *notification_data, uint16_t data_len)
1311 {
1312     struct bgp_data_notification data;
1313     uint16_t len = 0;
1314
1315     data.error_code = code;
1316     len += sizeof(data.error_code);
1317
1318     data.error_subcode = subcode;
1319     len += sizeof(data.error_code);
1320
1321     memcpy(data.data, notification_data, data_len);
1322     len += data_len;
1323
1324     memset(peer->outbuf->packet.header.marker, 0xff,
1325         sizeof(peer->outbuf->packet.header.marker));
1326
1327     peer->outbuf->packet.header.type = BGP_MSG_NOTIFICATION;
1328     peer->outbuf->packet.header.len =
1329         htons(sizeof(peer->outbuf->packet.header) + len);
1330
1331     memcpy(peer->outbuf->packet.data, &data, len);
1332
1333     peer->outbuf->done = 0;
1334     peer->next_state = code == BGP_ERR_CEASE ? Disabled : Idle;
1335
1336     /* we're dying; ignore any pending input */
1337     peer->inbuf->packet.header.len = 0;
1338     peer->inbuf->done = 0;
1339
1340     return bgp_write(peer);
1341 }